CCTV ATM Robbery Detection Dataset: Gun & Knife

Standard security models often confuse a friendly close-proximity interaction with an active armed robbery. This dataset is built to close that gap. Every frame is rendered from a realistic high-angle ATM surveillance perspective and annotated with four classes — offender, victim, gun, and knife — so models can learn not just what weapon is present, but who is wielding it and who is being threatened. Scenes cover varied lighting (bright daylight to low-light nocturnal), exterior brick walls, vestibules, drive-up kiosks, and standard street-corner ATM placements. Each image carries CCTV degradation realism: compression artifacts, sensor noise, and the desaturated color profile of real surveillance hardware. 100% synthetic — no real people, no biometrics, no GDPR exposure. The open-source sample includes 240 images on Kaggle; the full package contains 3,000 images covering a balanced mix of handgun and knife robbery scenarios across multiple ATM environments.